Study of 13,867 AI Disclosure Statements Finds Policies Vague and Practices Misaligned
danish037 · x · 2026-08-25
A paper accepted to EMNLP 2026 Findings, Expectations and Practices around AI Disclosure in CS Research, examines AI-use disclosure in CS research:
- Policies: Top CS venues widely require AI disclosure, but policies remain highly under-specified, with many barely changed since 2023.
- Researcher expectations (N=109 survey): Disclosures are deemed most necessary for research-design tasks (proposing hypotheses, generating synthetic data) and when human involvement is low; least necessary for writing and reporting.
- Practice: Analyzing 13,867 disclosure statements from EMNLP 2025 and ICLR 2026 reveals a large disconnect—writing assistance, considered least necessary to disclose, is the most frequently disclosed.
- Surprise finding: One disclosure statement was copied verbatim in 95 distinct ICLR 2026 papers; shared authors or a common university don't explain it.
The paper closes with recommendations to align disclosure policies and practices with expectations.
More from Safety
- Anthropic red team's first-ever model "abliteration" cut GLM-5.3 refusals from 90% to 3% for ~$4,400 — cephaloform · 2026-10-16
- Anthropic grants METR employee-like access as independent safety evaluators go public — AnthropicAI · 2026-10-16
- APEX-Harm benchmark: script-based injection attacks easily compromise agents in shared workspaces — KAUST · 2026-10-16
- Residual-stream forecasts predict agent violations before they happen, USC study shows — USC · 2026-10-16
- AI safety drama: risk research called a 'company effort,' team lead responds — logangraham · 2026-10-16
- Brave's Leo adds end-to-end encrypted chat with GLM via attested TEE — gnukeith · 2026-10-16